Simple package for Bayesian model comparison.
Given samples from a posterior distribution inferred under some default prior, compute the Bayes factor or odds in favour of a new prior model.
pip install popodds
The package consists of the ModelComparison
class to compute Bayes factors, and a wrapper function log_odds
for simplicity.
The computation only requires a few ingredients:
model
a new prior model or samples from it,prior
the original parameter estimation prior or samples from itsamples
samples from a parameter estimation run.
Optional:
model_bounds
parameter bounds for the new prior model,prior_bounds
parameter bounds for the original prior model,log
compute probability densities in log space,prior_odds
odds between the prior models, which defaults to unity,second_model
model to compute odds against instead of prior,second_bounds
parameter bounds for the second model,detectable
compare between detectable rather than intrinsic populations.